1. Patterns for Decoupling Data Structures and Algorithms or How visitors can help you grow! Stephen Wong, Oberlin College Dung “Zung” Nguyen, Pepperdine University

2. Where were we anyway? *Abstracting the structure and behavior of a linear recursive structure (“LRS” or “list”): *Empty vs Non-Empty States *State Transitions *State Design Pattern *Single LRS class that changes states. *Intelligent states that can handle themselves. First, the invariant behaviors…

3. UML Diagram of a LRS The public listThe abstract state The empty state The non-empty state

4. Object-Oriented Data Structures Represent the pure behavior of the structure of the data, independent of the data itself. Let’s take a look when we add the variant behaviors…

5. UML Diagram of a LRS Variant behaviors declared and accessed here Passed through this interface Executed here Or here

6. What’s wrong with this picture? *Why are the variant behaviors coded together with the invariant behaviors? *But aren’t the variant behaviors still the domain of the data structure object? *But who can predict future behavioral needs? *We need extensibility. *We need to de-couple the variant algorithms from the data structure. Now let’s fix it!

7. The Issue of Extensibility *Modern software engineering is driven by a need to handle change: *Code will be modified many times. *New features will be added on short intervals. *The complexity of modern system demands adherence to strict protocols. *Proper abstraction is crucial. *Need public interfaces that allow for expansion.

8. Examples of Extensible Programs *Netscape’s plug-ins. *MS Office’s add-ins and MS Active-X. *Dynamic Link Libraries (DLL’s). *I/O drivers But none of these can handle multiple hosts with different needs…. All of these can handle multiple extensions... Why shouldn’t they??

9. Designing for the Unknown *Identify the variant and invariant behaviors. *Encapsulate the invariant behaviors into a class. *Add hooks to this class to define communication protocols to other classes. *Encapsulate the variant behaviors into classes that comply with the above protocols.

10. The Visitor Design Pattern *Visitors *Encapsulate the variant behaviors. *All conform to a set invocation interface. *Provide different methods for different hosts. *Hosts *Provide a consistent “hook” for visitors. *Each different host calls only its desired method in the visitor.

11. Framework Control *Algorithm handed to the data structure rather than the other way around. *The data structure incapsulates the invariant behaviors and the algorithms are variant behaviors being added to it.

12. Visitors with a Common Interface Conforming algorithms Fixed visitor interface presents methods for each host.

13. Abstracted Hosts result = visitor.method1(this, param);result = visitor.method2(this, param);result = visitor.method3(this, param);User class calls for variant behavior in its abstract host: result = _host.visit(visitor, param); Each host calls its desired method

14. Binary Tree Structure Using the State Pattern Invariant behaviors plus a hook for the variant behaviors.

15. Binary Tree Structure Visitors

16. What Visitors Bring *Algorithms encapsulated separately from the data structure. *Invariant behaviors in the data structure. *Variant behaviors in the visitors. *Unlimited future capabilities. *Multiple host capable: Can handle state- dependent algorithms. *Recursive and non-recursive capable.

17. There’s more to come... *Recursive indexing schemes for arbitrarily branched tree structures. *Building a binary search tree using visitors. But not today…...

18. Growing with Visitors Bring on the visitors, for they bring abilities never before imagined.